Legume cultivation has a highly positive impact on agri-food systems by increasing the availability of biologically fixed nitrogen, enhancing soil quality, promoting biodiversity, and mitigating the impact of weeds and pests (Karkanis et al., 2016; Nemecek et al., 2008). For European agriculture, greater legume cultivation would help reduce its significant deficit and reliance on imported high-protein feedstuff (which contributes to Amazon deforestation: Boerema et al., 2016) and meet the increasing industry demand for novel protein-rich foods (Lucas et al., 2015; van Loon et al., 2023). Pea (Pisum sativum L.) and white lupin (Lupinus albus L.) are promising cool-season grain legume crops for southern Europe. Compared to other pulses, pea has a higher yield and energy production, while lupin maximizes protein yield per unit area due to its outstanding seed protein content (Karkanis et al., 2016; Annicchiarico, 2008; Cernay et al., 2016). However, greater plant breeding effort is indispensable to reduce the yield gap with cereals and increase the economic sustainability of these crops (Rubiales et al., 2021).

Cool-season grain legumes are typically sown in autumn in mild-winter regions and in late winter or early spring in cold-prone regions of Europe. The changing climate is expected to expand the autumn sowing the northwards, allowing crops to benefit from milder winters and escape the increasing risk of terminal drought through earlier crop maturity (Alessandri et al., 2014). Crop frost tolerance is a key breeding target in this context, not only to withstand low temperature stress in cold regions but also because sudden frost events following mild-temperature periods may produce high winter plant mortality due to insufficient cold acclimation (Annicchiarico and Iannucci, 2007). Various stresses may concurrently affect winter survival, including frost, waterlogging, and fungal pathogens (Bélanger et al., 2006). However, frost has prominent importance and can be faced by plants through frost avoidance and frost tolerance mechanisms (Janská et al., 2010). Frost avoidance is based on delayed flowering (aimed to protect the very sensitive reproductive organs) (Maqbool et al., 2010; Siddique et al., 1999), which is primarily achieved through greater vernalization requirements in white lupin (Huyghe and Papineau, 1990) and by photoperiodic control and/or high growing degree days requirements in pea (Summerfield and Roberts, 1988; Lejeune-Hénaut et al., 1999; Weller et al., 1997). In target regions possibly subjected to both low winter temperatures and terminal drought, late flowering and crop maturity may ensure frost avoidance and a higher yield in cold, relatively moisture-favorable years while being associated with greater drought susceptibility and lower yield in relatively mild, drought-prone years (Annicchiarico et al., 2011, 2019). For plant breeders, this dilemma can be coped with by selecting materials with intermediate flowering times but intrinsic frost and drought tolerance. Intrinsic drought tolerance could be expressed by a positive deviation from the genotype yield expected according to its onset of flowering (Annicchiarico et al., 2017; Pecetti et al., 2023). Intrinsic frost tolerance of cold-acclimated plants could likewise be expressed by a positive deviation from the genotype winter plant survival expected according to its onset of flowering. The frost tolerance mechanisms of cool-season grain legumes are based on physiological modifications to prevent or resist intracellular ice formation (Bourion et al., 2003), such as decreased shoot water content during cold acclimation (Sallam et al., 2015), increased cell membrane stability through changes in the lipid-to-protein ratio and the membrane lipid unsaturation level (Arbaoui and Link, 2008), and accumulation of osmoprotectant compounds such as proline, glycine betaine, mannitol, sucrose, raffinose, stachyose, and specific proteins that protect against dehydration (Link et al., 2010).

Breeding for improved frost tolerance under field conditions is complicated by the wide and increasing climate variation across years, which reduces the applicability, efficiency, and replicability of the selection (Avia et al., 2013). A reliable assessment of frost tolerance in controlled environments could overcome these limitations and allow, in addition, for off-season selections (Stoddard et al., 2006). Its assessment on seedlings rules out any effect of flowering time and focuses, therefore, on intrinsic frost tolerance. The assessment requires a period of cold acclimation (hardening) above 0°C temperature, the duration of which increases the frost tolerance of relatively winter-hardy material (Swensen and Murray, 1983). Hardening was performed at 4°C over 2 to 4 weeks in most freezing tolerance studies on pea (Auld et al., 1983; Murray et al., 1988; Meyer and Badaruddin, 2001; Homer et al., 2016). Subsequently, slow cooling toward the stress temperature is essential to ensure sufficient time for water redistribution, with a cooling rate not exceeding 2°C/h (Murray et al., 1988). An accurate assessment of mortality can only be made after a minimum recovery period of 3 weeks under favorable temperatures (Murray et al., 1988). Besides plant mortality, genotype frost tolerance could also be expressed by a visual score based on the amount of necrotic areas and other traits (Ali et al., 2016; Beji et al., 2020). The assessment of the genotype lethal temperature 50 (LT₅₀), i.e., the freezing temperature corresponding to 50% of mortality, requires the evaluation of plant mortality across a set of freezing temperatures and may, therefore, be operationally less adequate than the evaluation of plant mortality at just one optimal freezing temperature when assessing frost tolerance in large numbers of genotypes as in selection trials (Waalen et al., 2011). Such optimal temperature should ensure the maximization of genotype variation for plant mortality, and may approach the genotype mean value of LT₅₀ in studies including a sample of genotypes representative of the crop frost tolerance variation. Various studies suggested that this temperature may fall in the range of −7°C to −9°C (Auld et al., 1983; Swensen and Murray, 1983; Murray et al., 1988; Cousin et al., 1993; Homer et al., 2016) for pea based on small sets of genotypes mostly selected several decades ago, whereas no information is available for white lupin. For pea, an official frost tolerance evaluation test prescribes the assessment of candidate varieties at −8°C freezing temperature (Wery et al., 1994).

We recently established an easy-to-build, high-throughput phenotyping platform for frost tolerance assessment represented by a 13.6 m² growth freezing chamber with programmable temperature to be used for the selection and genomic prediction of frost tolerance in cool-season grain legumes. This study assessed plant mortality, LT₅₀ values, and the biomass injury visual score of 11 genotypes of pea and 11 of white lupin encompassing a wide range of winter mortality in earlier field trials in northern Italy, with the objectives of (i) optimizing the frost tolerance platform with respect to optimal freezing temperatures for each species and (ii) verifying the consistency of genotype plant mortality responses across platform and field conditions.

On average, pea exhibited greater frost tolerance than white lupin as indicated by the lower LT₅₀ (−12.8 versus −11.0°C; P < 0.01) and lower plant mortality at the lowest freezing temperature (0.50 versus 0.91; P < 0.01) in the analysis of variance-based species comparison. Within pea, the genotype values of LT₅₀ ranged from −14.5°C for the breeding line KI_L38 to −11.6°C for the cultivar Kaspa ( Table 2 ). The pea genotype plant mortality values showed no mortality at −7°C freezing temperature and did not differ significantly (P > 0.05) at −9°C ( Figure 1 ). They displayed significant differences (P < 0.01) at lower temperatures and achieved the largest variation, ranging from 0.11 to 0.83, at −13°C ( Figure 1 , Table 2 ). The biomass injury visual score of the pea genotypes decreased with increasing freezing temperature but exhibited significant (P < 0.01) and similar extents of overall genotype variation across all freezing temperatures ( Figure 1 ). The injury score values showed a high correlation (r ≥ 0.77, P < 0.01) across the four freezing temperatures ( Supplementary Table S2 ). The high correlation (P < 0.001) of genotype mortality at −13°C with biomass injury score for the same temperature (r = 0.98) and LT₅₀ (r = 0.91), and that between genotype values of the last two traits (r = 0.92), indicated a strong consistency between the main indicators of pea genotype frost tolerance.

Table 2 reports the results of pea genotype comparisons relative to LT₅₀, plant mortality at the two lowest freezing temperatures (those displaying a significant genotype variation), and the biomass injury score for the highest and lowest temperatures (i.e., the extreme temperature range). In general, the pea genotype mean separation was more sensitive for LT₅₀ (where KI_L38 outperformed any other genotype at P < 0.05) than for the other traits ( Table 2 ). All genotypes assigned to the high winter hardiness class on the basis of field observation exhibited low to fairly low values of LT₅₀, plant mortality, and injury score, while all genotypes assigned to the low winter hardiness class showed fairly high to high values of these traits ( Table 2 ). However, two genotypes in the intermediate winter hardiness class, namely, Dove and Kaspa, exhibited high and low frost tolerance, respectively, according to all traits. The highly winter-hardy genotype Champagne displayed high, but not outstanding, frost tolerance.

The white lupin genotypes displayed LT₅₀ values ranging from −12.0°C for the Greek landrace GR56 to −10.0°C for the Egyptian landrace Egypte11 and the Portuguese landrace E80. We observed no lupin genotype plant mortality at -7°C, and significant (P < 0.01) plant mortality variation only at −11°C and −13°C freezing temperatures ( Figure 2 ). However, the widest variation for genotype mortality, in the range of 0.26 to 0.88, took place at −11°C in this species ( Figure 2 , Table 3 ). The variation for the biomass injury score achieved significance (P < 0.01) only for the two intermediate temperatures while being flattened toward low values at −7°C and high values at −13°C ( Figure 2 ). A high consistency among major indicators of genotype frost tolerance was observed also in white lupin according to correlations (P < 0.001) of genotype mortality at −11°C with the biomass injury score for the same temperature (r = 0.97) and LT₅₀ (r = 0.94), or that between values of the last two traits (r = 0.91). High correlations were also observed for other indicators of frost tolerance ( Supplementary Table S3 ).

The results for major indicators of frost tolerance reported for each white lupin genotype in Table 3 indicated also for this species a good consistency between frost tolerance in the phenotyping platform and winter hardiness based on field observations. The five genotypes belonging to the high winter hardiness class were the top-ranking ones for frost tolerance according to LT₅₀ values, plant mortality, or the biomass injury score for −11°C, whereas three genotypes in the low winter hardiness class out of four were bottom-ranking for all of these frost tolerance indicators. The only inconsistencies were represented by the genotype E80, which was susceptible to frost while belonging to the intermediate winter hardiness class, and the breeding line PLI-P3, which displayed intermediate frost tolerance while being assigned to the low winter hardiness class. In white lupin, too, LT₅₀ exhibited more sensitive genotype mean separation than plant mortality or the injury score ( Table 3 ).

On average, pea exhibited greater frost tolerance than white lupin in this study. This result agrees with the greater average winter plant survival of pea relative to white lupin in a field-based assessment of a large number of varieties across climatically contrasting Italian environments (Annicchiarico and Iannucci, 2007). In general, however, pea is credited an intermediate winter hardiness among the cool-season grain legumes, being considered less winter-hardy than faba bean or lentil and more winter-hardy than chick pea (Murray et al., 1988).

Our results indicated a good consistency between major indicators of genotype frost tolerance observed in the phenotyping platform, namely, LT₅₀, plant mortality at the freezing temperature that maximized the genotype variation, and biomass injury score for the same temperature or a slightly higher one. LT₅₀ exhibited a more sensitive genotype mean separation than the other indicators. This characteristic, however, requires multiple freezing temperatures (four in our study), making it less suitable for evaluating large genotype numbers than the frost tolerance assessment based on one optimal freezing temperature (i.e., the one that maximizes the genotype variation for plant mortality). The optimal freezing temperature differed for the two species according to our results, being about −13°C for pea and −11°C for white lupin. When used for evaluation at an optimal freezing temperature, our platform could accommodate up to 216 genotypes in each of several evaluation cycles (each cycle acting as a replicate), using experimental units (replicates) of 10 plants each as in this study (or 144 genotypes, using experimental units of 15 plants). Our results suggest that the biomass injury score may concur to the frost tolerance evaluation along with plant mortality or act as the only frost tolerance indicator in case the platform included more genotypes per evaluation cycle with less plants per experimental unit (e.g., 432 genotypes with five plants per replicate), a situation that makes the estimation of plant mortality less reliable. A similar score was adopted by Beji et al. (2020) in a pea experiment including four plants per replicate, and is frequently adopted in other grain legumes under similar circumstances (Arbaoui and Link, 2008). Work by Humplík et al. (2015) suggests that the biomass injury assessment of the individual plants could be automated by image analysis, albeit hardly with large time savings and with a need for placing plants into individual pots.

The optimal freezing temperature for pea plant mortality at −13°C contrasts with earlier results by Auld et al. (1983), Swensen and Murray (1983), Murray et al. (1988), Cousin et al. (1993), and Homer et al. (2016), which suggested an optimal temperature in the range of −7°C to −9°C. The contrast is even greater when considering that most of these studies adopted a longer hardening period than our study. The improved frost tolerance of the current, recently bred germplasm sample (breeding lines or commercial cultivars), along with possible differences in the evaluation protocols, may partly account for the currently lower optimal freezing temperature. For example, differences in substrate type and drainage may affect plant mortality (Russell et al., 1978). Irrigation during the hardening period (as contemplated in some earlier studies) could increase ice formation and cause mechanical damage to the roots. Our slower thawing (1°C/h) relative to some early studies could be less damaging to plants (Gusta and Fowler, 1977). Other possibly different factors may include the plant growth and development stage before hardening, the duration of the frost treatment, and the length of the regrowth period. Although the frost tolerance evaluation of pea germplasm collections at −8°C is quite frequent (Wery et al., 1994; Dumont et al., 2009; Beji et al., 2020), Prieur and Cousin (1978) suggested the selection of frost-tolerant pea germplasm by a set of freezing cycles ultimately achieving −12°C.

No prior assessment of genotype frost tolerance variation and optimal temperature for frost tolerance evaluation based on LT₅₀ for plant mortality was available for white lupin. A study based on frost-induced leaf damage of cultivars and accessions estimated by chlorophyll fluorescence indicated an average value of −9.5°C for LT₅₀, estimated as 50% of damaged leaves after a long hardening period (42 days at 8°C/2°C day/night temperature) (Hoffmann-Bahnsen and Herzog, 2001). In contrast, Papineau and Huyghe (1992) proposed to assess white lupin frost tolerance at −16°C freezing temperature after a 3-week hardening period at −4°C.

The observed good consistency between platform-based frost tolerance and field-based winter hardiness of pea and white lupin genotypes has practical importance for the exploitation of artificial screening results for these species. Correlations for pea plant mortality across field and growth chamber assessments were close to 0.7 in Homer et al. (2016) and in the range of 0.5–0.6 in Auld et al. (1983). Correlations close to 0.5 have been reported for other legume species such as faba bean (Arbaoui et al., 2008) and red clover (Zanotto et al., 2021). As anticipated, one cannot expect a very high consistency between platform-based and field-based plant mortality in grain legumes because the latter depends not only on intrinsic frost tolerance but also on frost avoidance through a delayed onset of flowering. Other factors may influence the genotype variation for field plant survival, such as greater tolerance to diseases whose attack is favored by frost damage, such as Ascochyta spp. for pea (Maufras, 1997), and a different susceptibility to imbibitional chilling of the germinating seed due to seed coat variation for rapidity of imbibition (Wery et al., 1994). Onset of flowering may actually explain the response of the lupin breeding line PLI-P3, which featured moderate frost tolerance according to the freezing test while belonging to the low winter hardiness class according to field observations. The high winter mortality under field conditions of this line was associated with extreme earliness of flowering in Annicchiarico et al. (2011) (where this line is coded as P3), a feature that would definitely increase its sensitivity to frost because of the early differentiation of the floral apex (Huyghe and Papineau, 1990). The currently good but not outstanding intrinsic frost tolerance of the pea landrace Champagne ( Table 2 ) in spite of its reportedly extreme field-based winter survival (Prieur and Cousin, 1978; Dumont et al., 2011) may be accounted for by frost escape under field conditions via delayed flowering caused by possession of the Hr (high response to photoperiod) gene (Dumont et al., 2009). Indeed the Hr gene reportedly co-segregated with the most important quantitative trait loci (QTL) for frost tolerance (Lejeune-Hénaut et al., 2008). Anyway, the reliability of our genotype classifications for winter hardiness suffered from the limited field-based evaluation it was based upon. For example, the pea variety Dove, here classified as intermediate for winter hardiness while showing high frost tolerance according to freezing test results, exhibited moderately high frost tolerance across various cold-prone agricultural environments of France (UNIP-ITCF, 2001).

In conclusion, our results encourage the use of high-throughput phenotyping platforms such as the current one for the assessment of pea or white lupin frost tolerance aimed at plant breeding, molecular studies for detection of QTL (e.g., Beji et al., 2020) and/or definition of genome-enabled prediction models, or for investigation of physiological mechanisms regulating frost tolerance (e.g., Dumont et al., 2009).

The assessment under artificial conditions could overcome the increasing unpredictability of field-based evaluations. In addition, its focus on intrinsic frost tolerance (as implied by the evaluation of young plants that lack any differentiation of reproductive organs) facilitates the combination of cold tolerance and drought tolerance characteristics unrelated to flowering time in novel varieties featuring greater yield stability and adaptation to the increasingly variable climate conditions. Indeed our results for PLI-P3 and Champagne confirm that intrinsic frost tolerance is not necessarily related to plant mortality under field conditions for pea or white lupin genotypes, and a similar response was observed for a few faba bean genotypes (Arbaoui et al., 2008). For pea, a genomic selection model for intrinsic drought tolerance proved capable of producing material with a similar flowering time but with increased yielding ability under severe drought relative to its genetic base (Annicchiarico et al., 2020), while a similar model is awaiting exploitation for white lupin (Pecetti et al., 2023).